Hacker News Reader

Comment by chpatrick | Hacker News Reader

It's 1.08s on my computer for one line of Python, which is respectable:

  python2 -m timeit -v -n 1 -s "import numpy" "numpy.memmap('1_gb_file', mode='r').sum()"                                                                                    
  raw times: 1.08 1.09 1.08

View on HN

Isn't a lot of numpy implemented in C?

Sure, but the point is that you can get some pretty nice speeds without writing any C yourself.

But only for operations that someone else has written the C code for you. If you come up with some new algorithm that cannot be reduced to operations already implemented in C, then you would need to write the fast C implementation yourself.

Python is implemented in C as well and it's still a dynamic language, which was the point here, i guess.

There is a large difference between "implemented in C" (presumably talking about the runtime being implemented in C) and "Python that calls C code". If the implementation of Python is implemented in C as an interpreter#, then the interpreter is almost sure to be slower than whatever language/technology the interpreter was implemented with/on top of. This is pretty orthogonal to "Python that is implemented in <whatever>, and is doing most of the calculations via calling a C library". If you have a Python program that computes over megabytes of data, but it is calling a C library to do the heavy lifting, then the performance of Python itself is probably going to be negligible if 99.9% of the time is spent on the computation in the C code.

# Though could be a VM or something.

So what? Do you live in an alternative universe where you cannot use libraries implemented in C? Then you also cannot use, for example, Node.js (based on libuv).

Where is the sum?

Reply on news.ycombinator.com